The Freedom to Think Further

AI can make it easier to stop thinking. Bruce and I found ourselves wondering whether its greater possibility is exactly the opposite: giving us the freedom to think much further than we could before.

I had been reading three of Bruce’s recent pieces. One was about social care, another asked what might happen if artificial intelligence became more competent than humans, and the third looked at the future of leadership. They appeared to be three separate subjects, but I could see the same question underneath all of them.

AI may be able to clarify what should be done, but politicians still have to decide what society is prepared to pay for and live with. AI may become so competent that the greater danger is not that it takes power from us, but that we voluntarily hand over our judgement because doing so becomes easier. And if machines become better at optimisation, the future leader may need something different: enough understanding, authority and courage to tell the optimiser when it is wrong.

I brought those three thoughts to Bruce.

As usual, he complicated them.

A clearer answer does not remove the decision.

We started with social care. Bruce had put the problem through several AI systems, compared the answers and looked for areas of agreement and disagreement.

So I asked him what seemed an obvious question. If AI can give politicians a much clearer picture of the available choices and their consequences, shouldn’t political decision-making become easier?

“In theory, it gets easier,” Bruce said. “But in practice, it’s probably going to get harder.”

That distinction matters.

AI can help expose the choices. It can compare different approaches, identify contradictions, challenge assumptions and show consequences much faster than traditional methods. What it cannot do is make the human interests surrounding those choices disappear.

People still have positions to protect. Institutions still have territory. Governments still face voters. Organisations still worry about losing control.

Bruce had already experienced some of this himself. He had sent his AI-assisted analysis into parts of the wider discussion and found himself confronting what he called “institutional paranoia about control”.

The technology may clarify the problem. It does not remove the politics.

In fact, perhaps it does something more uncomfortable. It makes the real disagreement harder to disguise.

If ten different analyses all lead towards roughly the same difficult trade-off, we can no longer spend quite so much time pretending that another committee will eventually discover a painless answer.

At some point somebody has to choose.

That was my first distinction.

AI can improve the quality of the decision in front of us. It cannot take responsibility for living with the decision afterwards.

But Bruce’s way of using AI led us somewhere else.

Margaret Atwood was missing.

Bruce rarely asks one AI a question and accepts the answer. He uses several models, compares what they produce, looks for contradictions and sometimes asks one system to examine the answer of another.

“Don’t over-rely on one AI,” he told me.

Recently, he had asked several models about important female writers. The exercise produced answers, but afterwards Bruce realised that he had asked too narrow a question.

What, he wondered, might these writers tell us about the future and the age of AI?

So he asked again.

The answers changed.

Then Bruce noticed something else.

“What about Margaret Atwood?”

She had been largely absent.

That tiny moment seemed more important to me than a perfect AI-generated list would have been.

The machines had produced the material. Bruce noticed what was missing.

He changed the frame.

Then he asked again.

For years, one of the comforting things we have said about AI is that humans will remain important because humans ask the questions. Bruce sees that reassurance already becoming less comfortable.

Some AI systems now generate useful supplementary questions themselves.

“Well,” he said, “now AI is getting into the game of asking better questions. So you’ve got to ask even better questions.”

That sounds like a competition between human and machine.

I don’t think it has to be.

What Bruce was actually doing was using the machine to extend the territory available to his own curiosity. One answer created another question. One omission created another path. One model could challenge another. An idea that might once have required hours of searching could be tested in minutes.

The machine was not closing the discussion.

It was making the discussion larger.

And suddenly I could see the other side of the AI argument.

We talk constantly about what AI might take away from us.

But what if one of its biggest effects is to give ordinary individuals access to an intellectual freedom that previously belonged mainly to people with enormous amounts of time, specialist knowledge, research assistants or institutional resources?

That freedom is already tangible.

Bruce can sit at home, ask several systems the same question, compare the results, spot an omission, introduce Margaret Atwood, change the question and begin again.

The limiting factor is no longer simply how much information he can personally find.

Increasingly, the limiting factor is what he thinks to ask next.

Five minutes is not the answer.

Bruce gave me an even bigger example.

He had been exploring global AI governance. One of the models offered to draft an agreement.

So he let it.

Around fifteen pages appeared in about five minutes.

Then he gave the draft to other AI systems. He asked them to criticise it, compare approaches, identify weaknesses and try to improve it.

Within roughly an hour, he had something resembling a serious starting point.

Not an international agreement.

Not the final answer.

A starting point.

That distinction is important because this is where conversations about AI easily become silly. Nobody sensible is suggesting that five minutes with a language model should replace international negotiation.

But imagine what has changed.

The blank page has almost disappeared.

Instead of spending the first weeks of a project merely assembling what is already known, a person can begin with a substantial draft, attack it from several directions and spend more of their time asking what is wrong with it.

That changes the job.

It moves human effort away from producing the first answer and towards judging, challenging and improving what has been produced.

Bruce sometimes talks about getting “90% of the answer” surprisingly quickly.

I think the interesting part is what happens to the remaining ten per cent.

That is where context lives.

That is where interests collide.

That is where somebody notices Margaret Atwood is missing.

That is where a technically elegant answer encounters a human being who says, “Yes, but…”

And perhaps that remaining ten per cent becomes more important, not less, precisely because the first ninety per cent has become so cheap.

The strange connection.

This is where Bruce’s experiments begin to look slightly eccentric.

He has been asking AI to put writers and thinkers beside each other.

Victor Hugo and Ernest Hemingway.

Ayn Rand and William Golding.

Aldous Huxley and George Orwell.

Why?

Because sometimes the apparently strange comparison reveals the underlying question.

Very different writers can be wrestling with the same tensions: the individual and society, freedom and control, responsibility and dependence, competition and cooperation.

Bruce made a remark during our conversation that caught the whole thing rather neatly.

People have already been looking at these subjects.

“They haven’t made the connections.”

“And this,” he added, “is what AI can do.”

I called it revolutionary.

Bruce called it obvious.

Perhaps we were both right.

The individual ideas may be obvious. The new possibility is being able to collide them almost without cost.

If I ask what Victor Hugo and Hemingway might have to say to each other about responsibility and the answer is useless, very little has been lost.

Ask something else.

Put another two ideas together.

Change the angle.

Introduce an absurd assumption and see whether it exposes something useful.

Most strange questions will probably produce nothing.

But occasionally one will create a doorway.

That is a very different way of thinking about AI from asking it to write our emails so that we don’t have to.

One use removes intellectual effort.

The other creates room for more of it.

That, I think, is the distinction I had been looking for.

AI can be used to escape thinking, or it can be used to escape the limits around our thinking.

Those are not the same thing at all.

What do we do with the freedom?

At one point I suggested to Bruce that perhaps human beings are badly equipped for the amount of freedom this technology could eventually give us.

He corrected me.

“We’re well equipped for the wrong agendas.”

His example was competition.

In a competitive frame, almost every change begins with the same private calculation: what am I going to lose?

If AI can do something I do, what happens to my job? If another person gains influence, what happens to mine? If a new method works, what happens to the institution built around the old one?

Bruce contrasted that with cooperation, where the starting question changes: can we organise this so that more people gain from it?

That does not make conflict disappear. It simply changes the lens through which we first encounter change.

And it made our earlier discussion about leadership more concrete.

I had recently handed responsibility for an issue of The Pineapple to Janita. I wanted another generation and another perspective to shape it. AI remains part of our editorial process and can make recommendations about things such as sequencing.

But the machine does not get the final say.

Neither, if I am genuinely handing responsibility over, should I automatically get the final say.

That is harder than it sounds.

Handing somebody responsibility while quietly retaining control is not delegation. It is theatre.

Bruce then asked the uncomfortable human question behind my experiment.

Does the other person actually want the future I am imagining for them?

AI cannot answer that for her.

I cannot answer it for her either.

Suddenly the rather grand discussion about artificial intelligence had arrived somewhere very ordinary.

Two people working together.

One person learning when to step back.

Another deciding how far she wants to step forward.

A machine somewhere in the middle offering suggestions.

And humans still deciding what those suggestions mean.

Keeping the chair.

Towards the end of our conversation, Bruce described his current AI explorations as “intellectually indulgent”.

He is retired. He has left behind many of the meetings and administrative structures of working life, but he still has decades of experience, curiosity and an appetite for ideas. AI now gives him a way to wander very quickly from literature to governance, from leadership to social care, and from one question into another.

He does not know how long he will keep playing this particular game.

For the moment, he is enjoying it.

And I realised that Bruce himself is a useful answer to the concern with which I began.

Yes, AI can tempt us to surrender judgement. If every time we face a difficult question we type it into a machine and accept the first polished answer, we may gradually become very efficient at not thinking.

But that is a choice.

Bruce is making another one.

He uses AI to travel further than he could travel alone. He asks several systems. He makes odd connections. He challenges the answers. He notices what is missing. He changes the question. Sometimes he asks the machine to challenge itself.

The technology gives him more territory.

It does not decide where he should go.

That may be the real opportunity in front of us.

For most of history, our thinking has been constrained by access: which books we could find, which experts we knew, which disciplines we understood, which languages we spoke, and simply how many hours we had available.

AI begins to loosen some of those constraints.

That is an extraordinary freedom.

But freedom is not the same as wisdom.

The machine can take us much further across the landscape.

We still have to look around.

We still have to notice what is missing.

We still have to decide which direction matters.

And, however good the machine becomes, we should probably keep hold of the chair.